Shouldn't AI be able to take this one step further and just analyze the binary (of the samba server in this case) and create all kinds of interface specs from it?

Make the LLM operate the hypervisor VM so it can observe a binary as it executes to write specs for it?